Problem Description

This project aims to build an application that can recognize handwritting text, then converts them to digital text that can editable.

• Dataset: IAM Handwriting Database: https://fki.tic.heia-fr.ch/databases/iam-handwriting-database.

  This cannot be reachable, so we use the processed dataset from Kaggle: https://www.kaggle.com/datasets/nibinv23/iam-handwriting-word-database

• Method: Transformers

Installation

1. Clone project

Open your Command line, Terminal, Powershell and run this command:

git clone https://github.com/dtruong46me/handwritten-text-recognition.git

2. Download model

Since our model is too large, so you need to download manually.

• Base model: donut.pth (1.5 GB)

• Our model: trOCR.pth (1.4 GB)

Your downloaded model should be located in src/saved/your_model.pth

Example:

..
└───src
    ├───..
    ├───saved
    │   └───trOCR.pth
    │   └───donut.pth
    ├───..

In file src/utils/predict.py, handle your model you downloaded in the line 12

MODEL = "model_downloaded.pth"

3. Change directory to repository & execute main.py

Make sure that you are in this directory ..\handwritten-text-recognition> to execute this project

Install the libraries in requirements.txt:

pip install -r requirements.txt

python main.py

Contributions

Name	Student ID	Email
Vu Tuan Minh (C)	20210597	[email protected]
Le Tuan Anh	20214874	[email protected]
Bui Minh Quang	20214925	[email protected]
Dinh Nguyen Cong Quy	20214927	[email protected]
Phan Dinh Truong	20214937	[email protected]

References